Nebius monthly digest, January 2025

Now offering immediate access to 8 NVIDIA H200 GPUs via self-serviceNow offering immediate access to 8 NVIDIA H200 GPUs via self-service

You can now access 8 NVIDIA H200 GPUs, 16 NVIDIA H100 GPUs and 2 NVIDIA L40S GPUs via our console and start using them immediately!

Also important for our self-service users: the Explorer Tier of $1.5 per NVIDIA H100 GPU-hour, covering the first 1,000 hours each month, is extended until March.

Nebius AI Studio expanded with DeepSeek R1, text-to-image and moreNebius AI Studio expanded with DeepSeek R1, text-to-image and more

• Chinese AI New Year started early this time with DeepSeek R1 and V3, as noted by our guest author, Prof. Dr. Ivan Yamshchikov. Read his overview of these models and use them via Studio for a super-competitive price: $0.8 per 1M input tokens for R1.

• We’re excited to launch a text-to-image service with Flux Schnell, Flux Dev and Stable Diffusion XL models, delivering premium, production-ready image generation at a fraction of typical costs. Explore our docs for deeper insights.

• Since we’ve launched text-to-image, it’s the perfect time for a dedicated guide on prompts! Here’s how to create exceptional images with Flux on Studio. Bring your AI art game to the next level.

• We’ve also included vision models, a broader range of large language models, powerful embeddings and LoRA hosting. These updates are designed to help you build more versatile AI tools.

Nebius becomes a Reference Platform NVIDIA Cloud PartnerNebius becomes a Reference Platform NVIDIA Cloud Partner

Expanding our current partnership within the NVIDIA Partner Network as a Preferred-level Cloud Partner, we are proud to announce Nebius is now a Reference Platform NVIDIA Cloud Partner.

Our platform is built on NVIDIA Reference Architecture clusters, verified through adoption, strict adherence and validation of NVIDIA Cloud Partner Reference Architecture. These clusters align across the entire technology stack of hardware and software components. Learn more about NVIDIA Cloud Partner Reference Architectures here.

We released world’s first datasets for training software engineering agentsWe released world’s first datasets for training software engineering agents

More specifically, our AI R&D team uploaded two datasets to Hugging Face: nebius/SWE-bench-extra, containing 6,411 Issue-Pull Request pairs, and nebius/SWE-agent-trajectories, featuring 80,036 software engineering agent trajectories, where an agent attempts to solve these issues. To learn more about the release, check out our research blog post.

Meet SDKs for Go and PythonMeet SDKs for Go and Python

Our cloud SDK for Go is now generally available via GitHub. This major update in programmatic interfaces brings more flexibility for MLOps teams and AI developers. Of course, we’ve also supported the most popular language for AI and machine learning. To access the Python SDK, install it using a package manager or download the code from GitHub.

Customer success, backed by resultsCustomer success, backed by results

• We love to highlight consistent customer success, particularly in life sciences and healthcare, one of the key industries we serve. Did you know that innovators like Simulacra AI are transforming the quantum chemistry field by automatically generating high-precision datasets for molecular dynamics models at scale? Read more on this notable approach here.

• As you’re reading this, Wubble sets the mood with smooth tunes generated on Nebius. Wubble is a cutting-edge AI platform designed to empower businesses to generate high-quality, royalty-free music instantly, streamlining creative processes and unlocking limitless possibilities for marketing, advertising, podcasts, games, stores and more.

You can never have too many tutorialsYou can never have too many tutorials

• A new tutorial in the Managed Service for Kubernetes docs explains how to use Compute Cloud disks as persistent volumes for pods. Another Managed K8s feature covered by the docs is that new clusters now have three etcd key-value stores by default, ensuring high availability at no additional cost.

• New Nebius customers can now follow a guide about signing up for the platform to get up and run faster. And if you’d like to set up single sign-on (SSO) for your contributors to work with your resources in the cloud, we’ve prepared a tutorial that helps you add user accounts from identity federations to your tenant.

• While many focus on the modeling aspects during LLM fine-tuning, efficiently managing compute and experiment tracking is equally crucial. This guide demonstrates how to effectively leverage SkyPilot and MLflow, two powerful open-source tools, to orchestrate LLM fine-tuning.

• When adapting a large language model to your domain or specialized application, you want efficiency and a certain degree of simplicity. The setup of Managed K8s plus SkyPilot we’ve described here provides exactly that, albeit not as the only option. Meta Llama-3.1-8B is just an example here.